A watermark sharing scheme that generates high quality halftone images using Genetic Algorithm (GA) was extended to the method that decodes _NC₂ kinds of different watermark images from any two images selected from N generated ones. Although this method can clearly decode watermark images consisting of bold letters like Logo, it becomes difficult to detect watermark images consisting of fine letters or texture. In this paper, we modify the evaluation function on watermark appearance so that we can decode fine watermark images in this method. We verify the effectiveness of this method through computer simulation using some benchmark images.

View full abstract

In this paper, we focus on watermarking method that uses background texture for printing documents. We investigate a method that not only can produce perceptually natural background texture expressing gray levels of the background gray scale image but also can robustly decode watermark information even from copied and/or cropped partial document by using marker information embedded for synchronization. Main features of the proposed method are as follows: (i) The marker pattern expresses gray levels as well as watermark information. (ii) Gabor filters having different phase information are utilized to detect both marker and watermark information. (iii) Reliability coefficient is provided to each decoded watermark bit and the most plausible decoding is accomplished using a neuron model. We pursuit both production of natural background texture as well as robust watermark detection by the combined effects from the above ideas. We verify the performance of the proposed method by conducting some experiments on watermark detection against copying and cropping.

View full abstract

We propose a method for watermarking authentication information into a printed matter taking into account the appearance as a document with a dot pattern code based on an illustration. Since random dots are usually used for conventional dot pattern, a document with that dot pattern often gives readers strange feeling. On the contrary, we avoid appearing this kind of problems by using an image which is obtained through transforming an illustration into a dot pattern image as a dot pattern code. We developed two key techniques; one is a method for transforming from an illustration image to a dot pattern image which is suited for watermarking; the other one is a method for watermarking that allow us to reduce noise caused by errors in printing and scanning a document. We verified availability of the proposed method through an experiment on perception of subjects by appearance of printed images with a dot pattern code made by the proposed method.

View full abstract

Currently, the personal authentication technique based on string such as PIN and password are used in various situations. However, there is a risk to be guessed easily by others because many users are using own birth date, phone-number and address as the secret informations. Therefore, random numeric character and string unrelated personal data are recommended as the secret information. It makes users to have bigger burden of user memory. So, as a method to reduce the burden of memory the image authentication are studied.
Image authentication is a technique to use the images as the secret information. Existing image authentication is designed to select the images of secret information (Pass-Image) of directly technique, but Pass-Image of this technique has a high risk of leak to others by observation attack. Personal authentication technique to select some images at once by the numbers has been proposed to select Pass-Image indirectly in this research. This technique makes difficult to guess Pass-Image because there are some candidate images of Pass-Image.

View full abstract

This paper proposes a personal authentication method by writer verification using a sequential neural network which can reduce a sense of rejection and resistance problem exists in biometric authentication. The sequential neural network that can handle time series patterns applied to the field of offline writer verification is almost no example. This method can obtain the features such as character types, writing locus and input order of characters by using multiple handwritten characters treated as time series data. As feature extraction, we used a weighted direction index histogram method and a background propagation method. From the evaluation experiment at the three cases of illegal authentication, it is shown that the proposed personal authentication method is effective.

View full abstract

This paper proposes a dynamical visual learning method, which aims to identify person using multiple surveillance cameras sparsely distributed in space. In the proposed method, densely distributed multiple images are captured by interpolating the object's appearance in the sparsely distributed multiple images with a simple 3D face model, and generate two initial eigenspaces (an eigenspace for pose estimation and that for identification). In case another image is captured, the object's pose and name are estimated using the eigenspaces. The image is projected onto the 3D face model as texture information to improve the object's appearance, and the eigenspaces are regenerated. The discernment capability for person identification of the proposed method is shown by experimental results.

View full abstract

The silhouette image recognition technology can be adopted as a preprocessing for selecting candidates in security systems and image search systems. In this paper, we propose a method for recognizing silhouette images by representing them as graphs. First, a graph is obtained from an image. Then, dissimilarity between two graphs are calculated by the edit distance. Using the dissimilarities, the training images are embedded in the vector space using the multidimensional scaling. We propose a novel method for embedding a new graph into the vector space in order to recognize an unknown image. The experimental results with silhouette images confirm the effectiveness of the proposed method.

View full abstract

It is desired to recognize objects in images correctly. Recognition using structural feature has been studied. A single structure extracted by the medial axis transform from a digital silhouette image is not always an essential feature as a prototype that represents silhouettes in a category, because of noise and distortion. In this paper, we propose a method for recognizing silhouette images by obtaining an essential structure from the images of a category. The essential structure is defined as a weighted minimum common supergraph of graphs which are extracted from silhouette images. To show the validly of the proposed method, experiments are carried out for categorizing silhouette images.

View full abstract

Fast and accurate image retrieval is an essential technology for a surveillance system, and it can be used for many applications such as the robot vision and the content retrieval. Given an input image and a reference image, we focus on the problem of finding the most similar partial image to the reference image in the input image. The template matching is one of the fundamental techniques for the image retrieval. One of the drawbacks of the template matching is the high computational complexity. In this paper, we propose a method for speeding up the calculation of the template matching by approximating the reference image by a polynomial. An existing fast template matching method that uses the polynomial approximation can calculate similarities of all the partial images efficiently, whereas it is not adequate for calculating some designated partial images. In the proposed method, given an input image, the normalized cross correlations of the template and the designated partial images of the input image are calculated efficiently with the polynomial. Experimental results with various sizes of images show the effectiveness of the proposed method.

View full abstract

A real-time object tracking becomes more and more important in a wide variety of vehicle or surveillance camera systems. This paper focuses on implementation of a hardware engine for a real-time KLT Tracker (Kanade-Lucas-Tomasi) which is known as a robust object tracking algorithm. The proposed architecture includes stream data and parallel processing schemes for high performance. Furthermore, a memory I/F based on FIFO and image block buffer enable to reduce a memory access delay. Experimental results based on FPGA show that the proposed engine can attain 9ms @ 100MHz for a 640×480 image with 100 feature points.

View full abstract

In this paper, we propose a method of posture recognition of people by using 3D information. In this research, people's 3D information is acquired using FG vision sensor which has been developed. The state of the arm, the face and the body are detected from 3D data.

View full abstract

The concern of the intelligent transportation system rises and many driver support systems have been developed. Especially, systems that use images are paid attention recently, and various researches on recognizing and understanding the road environment have been done. Recognizing traffic lights are important for safety driving. If it is possible to detect and recognize a traffic light, it will give useful information to a driver to understand the road environment. In this paper, a fast method of detecting a traffic light in a scene image is proposed. By converting the color space from RGB to normalized RGB, some regions are selected as candidates of a traffic light. Then a method based on the Hough transform using the color information as well as the edge information is applied to detect an exact region. Experimental results using images including a traffic light taken by a digital camera through a windshield verifies the effectiveness of the proposed method.

View full abstract

The ditch of the tire has an important role for the safety running. Therefore, the depth of the ditch of the tire is checked without fail at the car inspection. However, this inspection is an easy method with a ditch gauge. On the other hand, the wear-out of the ditch is not the same because of the environmental condition. Therefore, it is necessary to measure all surroundings of the tire to the inspection. Authors constructed the ditch measurement system of the tire. The accuracy of the measurement data was examined. As a result, the effectiveness of the tire ditch measurement system was admitted.

View full abstract

At present, in making to be a code number by bank ATM, the numeral of 4 digits is mainly used. On the other hand, card falsification and unjust photographing in the code number input don't decrease. Then, the contrivance that the numeral arrangement of input screen of the code number is changed has also done it. And, the bank which establishes the vein authentication system using palm and finger which entered the practical stage recently increases. However, though the technique according to the biometric information of the vein certification is the high security, since card falsification and unjust photographing are difficult, on the other hand the introduction cost of it becomes considerable large amount of money. And, though it is the biometric information of fingerprint and vein of hand, there is a case in which it can not be used by the failure. And, the certification which contacts the equipment can not deny either the psychological resistance. In this study, security key entry system using infrared camera is developed. This proposal system allocates four shape patterns of hands as a code number, and it is made to be the security key. As a result of application experiment to the security key, it succeeded in the certification of the security key, and the effectiveness was shown.

View full abstract

Privacy protection can be achieved by unrecognizable method such as mosaicing. The method that unrecognizable objects can be reconstructed have been proposed. However, a method of reconstructed image authentication is necessary because reconstructed image is inadmissible. In this paper, we propose a method of image authentication by using encrypted stream by public key. This method can realize privacy protection and image authentication, because the image authentication does not require original image and privacy protect does not destroy its authenticity. Moreover, output data size increases little because the encrypted data for original image reconstruction is embedded into output image by using watermarking based on property of Huffman coding.

View full abstract

We propose a photogrammetric system based on visible light communications using a light marker as a reference mark. Our system automatically matches the light markers captured from multiple viewpoints. Each light blinks as a signal based on visible light communications. Using the signal, the automatic matching of the light markers is achieved. In addition, the automatic detection of the light markers is achieved by the rule of pulse position modulation and cyclic redundancy check. In our experimental results, the accuracy of the detection of the light areas and photogrammetry using light markers are discussed.

View full abstract

We developed a HD remote lecture system to link a science museum and a school. By compositing a teacher into image of the library picture or the simulator software realtime, he can get into the images and deliver a lecture. And teacher enables it to perform the change of library, operation of simulator software, and the writing into library picture by operating a chalk-type LED light. It is difficult to mix images by installing blue back on the museum, but by correction of the background subtraction, blue back and special equipment does not have to be prepared and thus a remote lecture can be realized easily from the museum. This developed system is used for the actual remote lectures and the practical performance is confirmed.

View full abstract

This paper proposes hybrid data representation schemes to improve the performance of a DCT-based steganographic method called StegErmelc¹⁾. Even though StegErmelc simultaneously achieves scalable carrier capacity, high embedding efficiency, and low SDR (stego detectable rate), SDR of StegErmelc increases sharply when the embedding rate is increased to more than 0.10bpc. SDR of StegErmelc also varies significantly when using different parameters. To suppress the sharp increase of SDR and to reduce the performance gap when using different parameters, we propose a new algorithm that keeps as many blocks that are operating in block mode as possible, and changes their encoding modes, one at a time, to gradually and accurately increase their carrier capacities. Experimental results show that, when embedding the same amount of information, the proposed method achieves lower SDR than that of StegErmelc and the performance gap of the proposed method is also found to be smaller than that of StegErmelc. The proposed method stays undetectable when embedding up to 0.125bpc. Basic performance of StegErmelc is also confirmed to be consistent regardless of the image database in use.

View full abstract

New variable cluster-size FM halftoning algorithm VACS (VAriable Cluster Screen) has been developed. This method could apply individual binarization processing after dividing input image data into 3 areas, shadow, intermediate and highlight. At intermediate area, Output-dependent Feedback Error Diffusion (Green-Noise Method) is applied as binarization algorithm, which enables to vary clustered-dot size in accordance with gain parameter. At highlight and shadow areas, homogeneous clustered-dots are created by deciding cell size, searching neighboring pixels already binarized. Each area is continuously and homogeneously connected introducing variable threshold T, which varies in proportion to the input image data, and also is adjusted cluster size at the boundary selecting suitable thresholds Ts and Th. As the result, homogeneous clustered-dots are obtained for all gradation levels. Tunable capability adjusting printing engine characteristics and high quality document image processing are feasible as VACS applications.

View full abstract

Subjective staircase illusion is a well-known binocular visual illusional phenomenon in psychology of vision and physics of vision. It is the phenomenon in which 3D staircase pattern appears, when a tilted dot matrix pattern is observed in binocular vision from the horizontal direction. Although the “nearest neighbour rule” in both-eyes' retina correspondence has recited in psycho-physical field as a fundamental hypothesis which causes this phenomenon, there is no verification of the rule. Then, in this research, the following verification was performed through the construction of a both-eyes' retina image computational model which imitated human's vision, and the computer experiment.
(1) Verification of whether staircase illusion occurs only from both eyes' retina image
(2) Verification of whether nearest neighbour rule works really correctly
About (2), because a more concrete and detailed rule expression was required to build a both-eyes' retina image computational model, “Nearest Neighbour Computational Rule” was newly formulized. Finally, while clarifying geometric structure of both eyes' retina image, it was verified that the staircase illusion occurs only from both-eyes' retina image of tilted dot matrix pattern.

View full abstract

The demand for the remote visualization has rising in order to construct remote collaborative work environment for working common issues by sharing visualization results of large-scale volmue data between remote places at real time, although the effective technique has not established. In this paper, we have proposed the remote visualization technique using particle-based volume rendering to solve the issues of the bottleneck by the composite processing in distributed visualization. In addition, we have implemented the parallel transmission by TCP as effective transmission of a large amount of data generated by dialogical processing via WAN. From the results of experiments on WAN, we have proved the parallel effect to display a result image of volume data by 512³ grids and the processing time become short by increasing the number of slave in server-side. Moreover, we have realized the reduction of data transmission time by increasing the number of TCP stream.

View full abstract

One of the methods for improving quality of full color image reproduction is an adaptive tone curve control. We proposed an adaptive learning tone correction curve method using a minimum color change method and multi functional layered networks with quantizer neurons (MFLN). The tone correction curve controls tone under the combined condition of constant color deference and constant RGB ratio (CDCR correction) and it can realize minimum change of hue and saturation in perceived color space. MFLN can select optimum tone correction curve according to a learning result of the relationship between optimum tone correction curve and input images. Tone curves of input images are corrected using the optimum tone correction curve and CDCR correction. As a result, high quality full color image reproduction can be obtained.

View full abstract

One of the biggest issues of an object recognition is the recognition with rotation invariance under a fluctuating noisy environment. We proposed a Temporal Pattern Recognition Network with Adaptive Segmentation of Quantizer Neuron Architecture (TASQA) and a φ-s transformation of shapes and applied them to object recognition.
The shape of the object is converted to a series of angles as a function of the circumference of the shape (φ-s data) and can be treated as a series of temporal patterns. TASQA consists of networks (ASQA network) with quantizer neurons, which can proliferate themselves and form the networks automatically during training, and a layer of neurons with self feedback (self feedback layer). The self feedback layer unifies the temporal recognition results of ASQA networks during a certain period defined by the time constant of self feedback and this function can realize the function of selective attention to certain areas of a series of temporal patterns. As a result, TASQA realizes rotation invariance in recognition and we obtained 100% recognition accuracy of 5 images with fluctuating noise taken by CCD camera.

View full abstract

In this paper, we propose a method to make digital content by using teacher tracking system. In order to make a digital content which includes the teaching action of the teacher and the writing on blackboard, we use two cameras to do the work. The first one is employed to track and record the teaching detail of the teacher during class (Teacher Tracking System, TTS), the other one is used to implement the acquisition system of writing on blackboard (ASOWOB). These two cameras are controlled by one computer, the position of the teacher is estimated by TTS, then send the command to ASOWOB to grab the image of writing on blackboard which the teacher is not in it. The digital content is built by composing the teaching detail of teacher and the image sequence of the writing on blackboard. Therefore, the student can study or review the whole lesson by it, the panorama of the whole blackboard also provide the overall sense of the class to the student. Our digital contents also can be published in internet, so the student can study in anywhere and anytime.

View full abstract

We present a technique for generating dendritic line (DL) images and apply it to image halftoning and an NPR for terrain images. The distance transform (DT) from edges extracted from an input image is computed. Branches are protracted from Poisson disk sampling (PDS) points to edges for producing DL images. Halftoning images are obtained by modulating the density of PDS points and the thickness of branches. For terrain images, a height field (HF) is generated from DL of ridges of DT and a colored texture image generated with a random dither is rendered with the HF.

View full abstract